SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2602.23614 · 2026

When Does Multimodal Learning Help in Healthcare? A Benchmark on EHR and Chest X-Ray Fusion

Chen Liu, Wenfang Yao, Kejing Yin, William Cheung, Jing Qin, Zijie Chen, Haizhou Xu, Yui Cheung

arXiv · PDF · Open in the Atlas

Code that ran

3 of 4 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (0 of them constructed an object rather than computing a result; 0 honoured the contract we drafted, 0 violated it and 0 have no contract verdict; 3 failures of our instrument; 1 no recorded run), measured 2026-11-22; bodies inline only under a permissive licence as GitHub detects it. Not a reproduction of the paper's results.

"Ran" means the function executed on a synthesized input and returned a value. It is not a reproduction of the paper's results. This record, dated, as JSON: /api/ran/2602.23614.json · badge: /api/ran/2602.23614.svg

Each repository's role is a label: a classifier's output recorded on the link, or, where no role is recorded, what the Papers with Code archive marked. The line under it names its basis. Neither says who wrote the code.

RepositoryRoleRan
jakeykj/CareBench canonical (classifier label)
classifier basis: census
3 ran · 1 no recorded run
FunctionStatusWhere it lives
create_ehr_encoder Ran
ran; our contract draft was wrong, not the code
jakeykj/CareBench/models/base/base_encoder.py
code served (permissive licence) · get_code("75abdad9153e475e")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned LSTMEncoder. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn

torch.manual_seed(0)

# The function signature is:
# def create_ehr_encoder(encoder_type, **kwargs):
#     return EHREncoderFactory.create_encoder(encoder_type, **kwargs)
#
# The encoder_type is used to select between 'lstm' and 'transformer'
# Both encoders need: input_size, num_classes, and optional hyperparameters

# Create minimal test data
batch_size = 2
seq_len = 5
input_size = 4
num_classes = 3
hidden_size = 8

# For LSTM encoder test
lstm_kwargs = {
    'encoder_type': 'lstm',
    'input_size': input_size,
    'num_classes': num_classes,
    'hidden_size': hidden_size,
    'num_layers': 2,
    'dropout': 0.1,
    'bidirectional': True,
}

# For Transformer encoder test (alternative)
transformer_kwargs = {
    'encoder_type': 'transformer',
    'input_size': input_size,
    'num_classes': num_classes,
    'd_model': hidden_size,
    'n_head': 2,
    'n_layers': 2,
    'dropout': 0.1,
    'max_len': 10,
}

# Use LSTM as the primary test case
ARGS = {
    'encoder_type': 'lstm',
    'input_size': input_size,
    'num_classes': num_classes,
    'hidden_size': hidden_size,
    'num_layers': 2,
    'dropout': 0.1,
    'bidirectional': True,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type LSTMEncoder — a LSTMEncoder output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a LSTMEncoder output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
load_yaml_config Ran
ran; our contract draft was wrong, not the code
jakeykj/CareBench/arguments.py
code served (permissive licence) · get_code("6a3ef7ec832c2fc1")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned dict. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import tempfile
import os
import yaml

# Create a temporary YAML file with some test configuration
temp_dir = tempfile.mkdtemp()
yaml_path = os.path.join(temp_dir, "test_config.yaml")

test_config = {
    "model": "test_model",
    "batch_size": 32,
    "learning_rate": 0.001,
    "layers": [64, 128, 256],
    "debug": True
}

with open(yaml_path, 'w') as f:
    yaml.dump(test_config, f)

ARGS = {
    "path": yaml_path
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
yaml — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type dict — a dict output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a dict output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
pad_temporal_data Ran
ran; our contract draft was wrong, not the code
jakeykj/CareBench/datasets/dataset.py
code served (permissive licence) · get_code("959fd46b219bad1f")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned dict. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import pandas as pd

torch.manual_seed(0)

# Construct a minimal batch for the collate function
# Each item in batch is a tuple of: (stay_ids, data, masks, cxr_imgs, has_cxr, labels, meta_attrs, idx, cxr_times)

batch_item_1 = (
    "stay_001",  # stay_ids
    torch.randn(3, 4),  # data: seq_len=3, features=4
    torch.ones(3, 4),  # masks: seq_len=3, features=4
    torch.randn(3, 224, 224),  # cxr_imgs
    1.0,  # has_cxr
    torch.tensor(1),  # labels
    {"age": 65, "gender": "M"},  # meta_attrs
    torch.tensor(0),  # idx
    100.5  # cxr_times
)

batch_item_2 = (
    "stay_002",  # stay_ids
    torch.randn(2, 4),  # data: seq_len=2, features=4 (different length)
    torch.ones(2, 4),  # masks: seq_len=2, features=4
    None,  # cxr_imgs (None case)
    0.0,  # has_cxr
    torch.tensor(0),  # labels
    {"age": 72, "gender": "F"},  # meta_attrs
    torch.tensor(1),  # idx
    50.2  # cxr_times
)

batch = [batch_item_1, batch_item_2]

ARGS = {
    "batch": batch
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
pandas, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type dict — a dict output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a dict output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
create_data_loaders Not yet run
unverified (harvested, no recorded run)
jakeykj/CareBench/datasets/dataset.py
code served (permissive licence) · get_code("baf76aa4175df88d")
How it ran

no invocation recorded.

Also present in other papers' code

Of 4 samples lifted from this paper's linked repositories, 4 carry a function identity and 0 do not. Of the 4 distinct functions with an identity, 0 are also present in other papers whose code we hold (0 with at least one paper dated earlier, 0 the same month, 0 later) and 4 are only here among the 52,977 papers whose code Syntology holds (as of 2026-09-26).

Only here means: among the 52,977 papers whose code Syntology holds (51,887 with lifted samples, 1,090 through their assemblies only), as of 2026-09-26, no other paper's lifted samples or current assembly bricks contain an identical function body after normalisation.

Presence, not direction: an identical function body in two papers' repositories does not say who wrote it, which came from which, or that either paper's method rests on it. Matching is exact after normalisation, so a renamed or edited function is not matched. Dates are when papers appeared, not when code was written. Only .py files the harvester lifted are covered; notebooks are not read. A new-style arXiv id YYMM.NNNN dates a paper to the month 20YY-MM; otherwise Paper.year dates it to the year; otherwise it is undated. Dates are when the paper appeared, not when the code was written. Read 2026-11-22.

Repositories linked to this paper

Some links come from the archived Papers with Code dataset (CC BY-SA 4.0): attribution and licence.

Abstract

Machine learning holds promise for advancing clinical decision support, yet it remains unclear when multimodal learning truly helps in practice, particularly under modality missingness and fairness constraints. In this work, we conduct a systematic benchmark of multimodal fusion between Electronic Health Records (EHR) and chest X-rays (CXR) on standardized cohorts from MIMIC-IV and MIMIC-CXR, aiming to answer four fundamental questions: when multimodal fusion improves clinical prediction, how different fusion strategies compare, how robust existing methods are to missing modalities, and whether multimodal models achieve algorithmic fairness. Our study reveals several key insights. Multimodal fusion improves performance when modalities are complete, with gains concentrating in diseases that require complementary information from both EHR and CXR. While cross-modal learning mechanisms capture clinically meaningful dependencies beyond simple concatenation, the rich temporal structure of EHR introduces strong modality imbalance that architectural complexity alone cannot overcome. Under realistic missingness, multimodal benefits rapidly degrade unless models are explicitly designed to handle incomplete inputs. Moreover, multimodal fusion does not inherently improve fairness, with subgroup disparities mainly arising from unequal sensitivity across demographic groups. To support reproducible and extensible evaluation, we further release a flexible benchmarking toolkit that enables plug-and-play integration of new models and datasets. Together, this work provides actionable guidance on when multimodal learning helps, when it fails, and why, laying the foundation for developing clinically deployable multimodal systems that are both effective and reliable. The open-source toolkit can be found at https://github.com/jakeykj/CareBench.

For agents

The same record, over MCP at https://syntology.ai/mcp:

get_harvested_code_for_paper("2602.23614")
get_code_for_paper("2602.23614")
have("2602.23614")

The run record, dated, one paper per request, free:

curl https://syntology.ai/api/ran/2602.23614.json

A badge for a README (the split and the date, never a ratio):

[![Syntology run record](https://syntology.ai/api/ran/2602.23614.svg)](https://syntology.ai/paper/2602.23614)

Connect an agent — have() is free.